Commit | Line | Data |
---|---|---|
b2441318 | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
5b99cd0e HC |
2 | #ifndef _LINUX_MM_TYPES_H |
3 | #define _LINUX_MM_TYPES_H | |
4 | ||
2e58f173 IM |
5 | #include <linux/mm_types_task.h> |
6 | ||
4f9a58d7 | 7 | #include <linux/auxvec.h> |
5b99cd0e HC |
8 | #include <linux/list.h> |
9 | #include <linux/spinlock.h> | |
c92ff1bd MS |
10 | #include <linux/rbtree.h> |
11 | #include <linux/rwsem.h> | |
12 | #include <linux/completion.h> | |
cddb8a5c | 13 | #include <linux/cpumask.h> |
d4b3b638 | 14 | #include <linux/uprobes.h> |
bbeae5b0 | 15 | #include <linux/page-flags-layout.h> |
ec8d7c14 | 16 | #include <linux/workqueue.h> |
2e58f173 | 17 | |
c92ff1bd | 18 | #include <asm/mmu.h> |
5b99cd0e | 19 | |
4f9a58d7 OH |
20 | #ifndef AT_VECTOR_SIZE_ARCH |
21 | #define AT_VECTOR_SIZE_ARCH 0 | |
22 | #endif | |
23 | #define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1)) | |
24 | ||
1c8f4220 SJ |
25 | typedef int vm_fault_t; |
26 | ||
5b99cd0e | 27 | struct address_space; |
1306a85a | 28 | struct mem_cgroup; |
133ff0ea | 29 | struct hmm; |
5b99cd0e HC |
30 | |
31 | /* | |
32 | * Each physical page in the system has a struct page associated with | |
33 | * it to keep track of whatever it is we are using the page for at the | |
34 | * moment. Note that we have no way to track which tasks are using | |
35 | * a page, though if it is a pagecache page, rmap structures can tell us | |
97b4a671 | 36 | * who is mapping it. |
be50015d | 37 | * |
97b4a671 MW |
38 | * If you allocate the page using alloc_pages(), you can use some of the |
39 | * space in struct page for your own purposes. The five words in the main | |
40 | * union are available, except for bit 0 of the first word which must be | |
41 | * kept clear. Many users use this word to store a pointer to an object | |
42 | * which is guaranteed to be aligned. If you use the same storage as | |
43 | * page->mapping, you must restore it to NULL before freeing the page. | |
be50015d | 44 | * |
97b4a671 MW |
45 | * If your page will not be mapped to userspace, you can also use the four |
46 | * bytes in the mapcount union, but you must call page_mapcount_reset() | |
47 | * before freeing it. | |
48 | * | |
49 | * If you want to use the refcount field, it must be used in such a way | |
50 | * that other CPUs temporarily incrementing and then decrementing the | |
51 | * refcount does not cause problems. On receiving the page from | |
52 | * alloc_pages(), the refcount will be positive. | |
53 | * | |
54 | * If you allocate pages of order > 0, you can use some of the fields | |
55 | * in each subpage, but you may need to restore some of their values | |
56 | * afterwards. | |
fc9bb8c7 | 57 | * |
4cf7c8bf MW |
58 | * SLUB uses cmpxchg_double() to atomically update its freelist and |
59 | * counters. That requires that freelist & counters be adjacent and | |
60 | * double-word aligned. We align all struct pages to double-word | |
61 | * boundaries, and ensure that 'freelist' is aligned within the | |
62 | * struct. | |
5b99cd0e | 63 | */ |
e20df2c6 MW |
64 | #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE |
65 | #define _struct_page_alignment __aligned(2 * sizeof(unsigned long)) | |
66 | #else | |
0dd4da5b | 67 | #define _struct_page_alignment |
7d27a04b | 68 | #endif |
e20df2c6 | 69 | |
5b99cd0e HC |
70 | struct page { |
71 | unsigned long flags; /* Atomic flags, some possibly | |
72 | * updated asynchronously */ | |
b7ccc7f8 | 73 | /* |
4da1984e MW |
74 | * Five words (20/40 bytes) are available in this union. |
75 | * WARNING: bit 0 of the first word is used for PageTail(). That | |
76 | * means the other users of this union MUST NOT use the bit to | |
b7ccc7f8 MW |
77 | * avoid collision and false-positive PageTail(). |
78 | */ | |
8456a648 | 79 | union { |
66a6ffd2 | 80 | struct { /* Page cache and anonymous pages */ |
4da1984e MW |
81 | /** |
82 | * @lru: Pageout list, eg. active_list protected by | |
83 | * zone_lru_lock. Sometimes used as a generic list | |
84 | * by the page owner. | |
85 | */ | |
86 | struct list_head lru; | |
66a6ffd2 MW |
87 | /* See page-flags.h for PAGE_MAPPING_FLAGS */ |
88 | struct address_space *mapping; | |
89 | pgoff_t index; /* Our offset within mapping. */ | |
90 | /** | |
91 | * @private: Mapping-private opaque data. | |
92 | * Usually used for buffer_heads if PagePrivate. | |
93 | * Used for swp_entry_t if PageSwapCache. | |
94 | * Indicates order in the buddy system if PageBuddy. | |
95 | */ | |
96 | unsigned long private; | |
97 | }; | |
c25fff71 JDB |
98 | struct { /* page_pool used by netstack */ |
99 | /** | |
100 | * @dma_addr: might require a 64-bit value even on | |
101 | * 32-bit architectures. | |
102 | */ | |
103 | dma_addr_t dma_addr; | |
104 | }; | |
66a6ffd2 | 105 | struct { /* slab, slob and slub */ |
4da1984e MW |
106 | union { |
107 | struct list_head slab_list; /* uses lru */ | |
108 | struct { /* Partial pages */ | |
109 | struct page *next; | |
110 | #ifdef CONFIG_64BIT | |
111 | int pages; /* Nr of pages left */ | |
112 | int pobjects; /* Approximate count */ | |
113 | #else | |
114 | short int pages; | |
115 | short int pobjects; | |
116 | #endif | |
117 | }; | |
118 | }; | |
66a6ffd2 MW |
119 | struct kmem_cache *slab_cache; /* not slob */ |
120 | /* Double-word boundary */ | |
121 | void *freelist; /* first free object */ | |
122 | union { | |
123 | void *s_mem; /* slab: first object */ | |
124 | unsigned long counters; /* SLUB */ | |
125 | struct { /* SLUB */ | |
126 | unsigned inuse:16; | |
127 | unsigned objects:15; | |
128 | unsigned frozen:1; | |
129 | }; | |
130 | }; | |
131 | }; | |
4da1984e MW |
132 | struct { /* Tail pages of compound page */ |
133 | unsigned long compound_head; /* Bit zero is set */ | |
134 | ||
135 | /* First tail page only */ | |
136 | unsigned char compound_dtor; | |
137 | unsigned char compound_order; | |
138 | atomic_t compound_mapcount; | |
139 | }; | |
140 | struct { /* Second tail page of compound page */ | |
141 | unsigned long _compound_pad_1; /* compound_head */ | |
142 | unsigned long _compound_pad_2; | |
143 | struct list_head deferred_list; | |
144 | }; | |
66a6ffd2 | 145 | struct { /* Page table pages */ |
4da1984e MW |
146 | unsigned long _pt_pad_1; /* compound_head */ |
147 | pgtable_t pmd_huge_pte; /* protected by page->ptl */ | |
66a6ffd2 | 148 | unsigned long _pt_pad_2; /* mapping */ |
4231aba0 NP |
149 | union { |
150 | struct mm_struct *pt_mm; /* x86 pgds only */ | |
151 | atomic_t pt_frag_refcount; /* powerpc */ | |
152 | }; | |
7d27a04b | 153 | #if ALLOC_SPLIT_PTLOCKS |
66a6ffd2 | 154 | spinlock_t *ptl; |
7d27a04b | 155 | #else |
66a6ffd2 | 156 | spinlock_t ptl; |
7d27a04b | 157 | #endif |
7d27a04b | 158 | }; |
50e7fbc3 MW |
159 | struct { /* ZONE_DEVICE pages */ |
160 | /** @pgmap: Points to the hosting device page map. */ | |
161 | struct dev_pagemap *pgmap; | |
162 | unsigned long hmm_data; | |
163 | unsigned long _zd_pad_1; /* uses mapping */ | |
164 | }; | |
4da1984e MW |
165 | |
166 | /** @rcu_head: You can use this to free a page by RCU. */ | |
167 | struct rcu_head rcu_head; | |
7d27a04b MW |
168 | }; |
169 | ||
b21999da MW |
170 | union { /* This union is 4 bytes in size. */ |
171 | /* | |
172 | * If the page can be mapped to userspace, encodes the number | |
173 | * of times this page is referenced by a page table. | |
174 | */ | |
175 | atomic_t _mapcount; | |
176 | ||
6e292b9b MW |
177 | /* |
178 | * If the page is neither PageSlab nor mappable to userspace, | |
179 | * the value stored here may help determine what this page | |
180 | * is used for. See page-flags.h for a list of page types | |
181 | * which are currently stored here. | |
182 | */ | |
183 | unsigned int page_type; | |
184 | ||
ca9c88c7 | 185 | unsigned int active; /* SLAB */ |
ca9c88c7 | 186 | int units; /* SLOB */ |
81819f0f | 187 | }; |
fc9bb8c7 | 188 | |
b21999da MW |
189 | /* Usage count. *DO NOT USE DIRECTLY*. See page_ref.h */ |
190 | atomic_t _refcount; | |
191 | ||
1306a85a JW |
192 | #ifdef CONFIG_MEMCG |
193 | struct mem_cgroup *mem_cgroup; | |
194 | #endif | |
195 | ||
5b99cd0e HC |
196 | /* |
197 | * On machines where all RAM is mapped into kernel address space, | |
198 | * we can simply calculate the virtual address. On machines with | |
199 | * highmem some memory is mapped into kernel virtual memory | |
200 | * dynamically, so we need a place to store that address. | |
201 | * Note that this field could be 16 bits on x86 ... ;) | |
202 | * | |
203 | * Architectures with slow multiplication can define | |
204 | * WANT_PAGE_VIRTUAL in asm/page.h | |
205 | */ | |
206 | #if defined(WANT_PAGE_VIRTUAL) | |
207 | void *virtual; /* Kernel virtual address (NULL if | |
208 | not kmapped, ie. highmem) */ | |
209 | #endif /* WANT_PAGE_VIRTUAL */ | |
dfec072e | 210 | |
90572890 PZ |
211 | #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS |
212 | int _last_cpupid; | |
57e0a030 | 213 | #endif |
e20df2c6 | 214 | } _struct_page_alignment; |
5b99cd0e | 215 | |
d1402fc7 LG |
216 | /* |
217 | * Used for sizing the vmemmap region on some architectures | |
218 | */ | |
219 | #define STRUCT_PAGE_MAX_SHIFT (order_base_2(sizeof(struct page))) | |
220 | ||
b63ae8ca AD |
221 | #define PAGE_FRAG_CACHE_MAX_SIZE __ALIGN_MASK(32768, ~PAGE_MASK) |
222 | #define PAGE_FRAG_CACHE_MAX_ORDER get_order(PAGE_FRAG_CACHE_MAX_SIZE) | |
223 | ||
224 | struct page_frag_cache { | |
225 | void * va; | |
226 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
227 | __u16 offset; | |
228 | __u16 size; | |
229 | #else | |
230 | __u32 offset; | |
231 | #endif | |
232 | /* we maintain a pagecount bias, so that we dont dirty cache line | |
0139aa7b | 233 | * containing page->_refcount every time we allocate a fragment. |
b63ae8ca AD |
234 | */ |
235 | unsigned int pagecnt_bias; | |
236 | bool pfmemalloc; | |
237 | }; | |
238 | ||
64b990d2 | 239 | typedef unsigned long vm_flags_t; |
ca16d140 | 240 | |
8feae131 DH |
241 | /* |
242 | * A region containing a mapping of a non-memory backed file under NOMMU | |
243 | * conditions. These are held in a global tree and are pinned by the VMAs that | |
244 | * map parts of them. | |
245 | */ | |
246 | struct vm_region { | |
247 | struct rb_node vm_rb; /* link in global region tree */ | |
ca16d140 | 248 | vm_flags_t vm_flags; /* VMA vm_flags */ |
8feae131 DH |
249 | unsigned long vm_start; /* start address of region */ |
250 | unsigned long vm_end; /* region initialised to here */ | |
dd8632a1 | 251 | unsigned long vm_top; /* region allocated to here */ |
8feae131 DH |
252 | unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */ |
253 | struct file *vm_file; /* the backing file or NULL */ | |
254 | ||
1e2ae599 | 255 | int vm_usage; /* region usage count (access under nommu_region_sem) */ |
cfe79c00 MF |
256 | bool vm_icache_flushed : 1; /* true if the icache has been flushed for |
257 | * this region */ | |
8feae131 DH |
258 | }; |
259 | ||
745f234b AA |
260 | #ifdef CONFIG_USERFAULTFD |
261 | #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, }) | |
262 | struct vm_userfaultfd_ctx { | |
263 | struct userfaultfd_ctx *ctx; | |
264 | }; | |
265 | #else /* CONFIG_USERFAULTFD */ | |
266 | #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {}) | |
267 | struct vm_userfaultfd_ctx {}; | |
268 | #endif /* CONFIG_USERFAULTFD */ | |
269 | ||
c92ff1bd MS |
270 | /* |
271 | * This struct defines a memory VMM memory area. There is one of these | |
272 | * per VM-area/task. A VM area is any part of the process virtual memory | |
273 | * space that has a special rule for the page-fault handlers (ie a shared | |
274 | * library, the executable area etc). | |
275 | */ | |
276 | struct vm_area_struct { | |
e4c6bfd2 RR |
277 | /* The first cache line has the info for VMA tree walking. */ |
278 | ||
c92ff1bd MS |
279 | unsigned long vm_start; /* Our start address within vm_mm. */ |
280 | unsigned long vm_end; /* The first byte after our end address | |
281 | within vm_mm. */ | |
282 | ||
283 | /* linked list of VM areas per task, sorted by address */ | |
297c5eee | 284 | struct vm_area_struct *vm_next, *vm_prev; |
c92ff1bd | 285 | |
c92ff1bd MS |
286 | struct rb_node vm_rb; |
287 | ||
d3737187 ML |
288 | /* |
289 | * Largest free memory gap in bytes to the left of this VMA. | |
290 | * Either between this VMA and vma->vm_prev, or between one of the | |
291 | * VMAs below us in the VMA rbtree and its ->vm_prev. This helps | |
292 | * get_unmapped_area find a free area of the right size. | |
293 | */ | |
294 | unsigned long rb_subtree_gap; | |
295 | ||
e4c6bfd2 RR |
296 | /* Second cache line starts here. */ |
297 | ||
298 | struct mm_struct *vm_mm; /* The address space we belong to. */ | |
299 | pgprot_t vm_page_prot; /* Access permissions of this VMA. */ | |
300 | unsigned long vm_flags; /* Flags, see mm.h. */ | |
301 | ||
c92ff1bd MS |
302 | /* |
303 | * For areas with an address space and backing store, | |
27ba0644 | 304 | * linkage into the address_space->i_mmap interval tree. |
c92ff1bd | 305 | */ |
ac51b934 KS |
306 | struct { |
307 | struct rb_node rb; | |
308 | unsigned long rb_subtree_last; | |
c92ff1bd MS |
309 | } shared; |
310 | ||
311 | /* | |
312 | * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma | |
313 | * list, after a COW of one of the file pages. A MAP_SHARED vma | |
314 | * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack | |
315 | * or brk vma (with NULL file) can only be in an anon_vma list. | |
316 | */ | |
5beb4930 RR |
317 | struct list_head anon_vma_chain; /* Serialized by mmap_sem & |
318 | * page_table_lock */ | |
c92ff1bd MS |
319 | struct anon_vma *anon_vma; /* Serialized by page_table_lock */ |
320 | ||
321 | /* Function pointers to deal with this struct. */ | |
f0f37e2f | 322 | const struct vm_operations_struct *vm_ops; |
c92ff1bd MS |
323 | |
324 | /* Information about our backing store: */ | |
325 | unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE | |
ea1754a0 | 326 | units */ |
c92ff1bd MS |
327 | struct file * vm_file; /* File we map to (can be NULL). */ |
328 | void * vm_private_data; /* was vm_pte (shared mem) */ | |
c92ff1bd | 329 | |
ec560175 | 330 | atomic_long_t swap_readahead_info; |
c92ff1bd | 331 | #ifndef CONFIG_MMU |
8feae131 | 332 | struct vm_region *vm_region; /* NOMMU mapping region */ |
c92ff1bd MS |
333 | #endif |
334 | #ifdef CONFIG_NUMA | |
335 | struct mempolicy *vm_policy; /* NUMA policy for the VMA */ | |
336 | #endif | |
745f234b | 337 | struct vm_userfaultfd_ctx vm_userfaultfd_ctx; |
3859a271 | 338 | } __randomize_layout; |
c92ff1bd | 339 | |
b564daf8 ON |
340 | struct core_thread { |
341 | struct task_struct *task; | |
342 | struct core_thread *next; | |
343 | }; | |
344 | ||
32ecb1f2 | 345 | struct core_state { |
c5f1cc8c | 346 | atomic_t nr_threads; |
b564daf8 | 347 | struct core_thread dumper; |
32ecb1f2 ON |
348 | struct completion startup; |
349 | }; | |
350 | ||
db446a08 | 351 | struct kioctx_table; |
c92ff1bd | 352 | struct mm_struct { |
c1a2f7f0 RR |
353 | struct { |
354 | struct vm_area_struct *mmap; /* list of VMAs */ | |
355 | struct rb_root mm_rb; | |
7a9cdebd | 356 | u64 vmacache_seqnum; /* per-thread vmacache */ |
efc1a3b1 | 357 | #ifdef CONFIG_MMU |
c1a2f7f0 | 358 | unsigned long (*get_unmapped_area) (struct file *filp, |
c92ff1bd MS |
359 | unsigned long addr, unsigned long len, |
360 | unsigned long pgoff, unsigned long flags); | |
efc1a3b1 | 361 | #endif |
c1a2f7f0 RR |
362 | unsigned long mmap_base; /* base of mmap area */ |
363 | unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */ | |
1b028f78 | 364 | #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES |
c1a2f7f0 RR |
365 | /* Base adresses for compatible mmap() */ |
366 | unsigned long mmap_compat_base; | |
367 | unsigned long mmap_compat_legacy_base; | |
1b028f78 | 368 | #endif |
c1a2f7f0 RR |
369 | unsigned long task_size; /* size of task vm space */ |
370 | unsigned long highest_vm_end; /* highest vma end address */ | |
371 | pgd_t * pgd; | |
372 | ||
373 | /** | |
374 | * @mm_users: The number of users including userspace. | |
375 | * | |
376 | * Use mmget()/mmget_not_zero()/mmput() to modify. When this | |
377 | * drops to 0 (i.e. when the task exits and there are no other | |
378 | * temporary reference holders), we also release a reference on | |
379 | * @mm_count (which may then free the &struct mm_struct if | |
380 | * @mm_count also drops to 0). | |
381 | */ | |
382 | atomic_t mm_users; | |
383 | ||
384 | /** | |
385 | * @mm_count: The number of references to &struct mm_struct | |
386 | * (@mm_users count as 1). | |
387 | * | |
388 | * Use mmgrab()/mmdrop() to modify. When this drops to 0, the | |
389 | * &struct mm_struct is freed. | |
390 | */ | |
391 | atomic_t mm_count; | |
b279ddc3 | 392 | |
c4812909 | 393 | #ifdef CONFIG_MMU |
c1a2f7f0 | 394 | atomic_long_t pgtables_bytes; /* PTE page table pages */ |
5a3fbef3 | 395 | #endif |
c1a2f7f0 | 396 | int map_count; /* number of VMAs */ |
481b4bb5 | 397 | |
c1a2f7f0 RR |
398 | spinlock_t page_table_lock; /* Protects page tables and some |
399 | * counters | |
400 | */ | |
401 | struct rw_semaphore mmap_sem; | |
c92ff1bd | 402 | |
c1a2f7f0 RR |
403 | struct list_head mmlist; /* List of maybe swapped mm's. These |
404 | * are globally strung together off | |
405 | * init_mm.mmlist, and are protected | |
406 | * by mmlist_lock | |
407 | */ | |
c92ff1bd | 408 | |
c92ff1bd | 409 | |
c1a2f7f0 RR |
410 | unsigned long hiwater_rss; /* High-watermark of RSS usage */ |
411 | unsigned long hiwater_vm; /* High-water virtual memory usage */ | |
c92ff1bd | 412 | |
c1a2f7f0 RR |
413 | unsigned long total_vm; /* Total pages mapped */ |
414 | unsigned long locked_vm; /* Pages that have PG_mlocked set */ | |
415 | unsigned long pinned_vm; /* Refcount permanently increased */ | |
416 | unsigned long data_vm; /* VM_WRITE & ~VM_SHARED & ~VM_STACK */ | |
417 | unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE & ~VM_STACK */ | |
418 | unsigned long stack_vm; /* VM_STACK */ | |
419 | unsigned long def_flags; | |
88aa7cc6 | 420 | |
c1a2f7f0 RR |
421 | spinlock_t arg_lock; /* protect the below fields */ |
422 | unsigned long start_code, end_code, start_data, end_data; | |
423 | unsigned long start_brk, brk, start_stack; | |
424 | unsigned long arg_start, arg_end, env_start, env_end; | |
c92ff1bd | 425 | |
c1a2f7f0 | 426 | unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */ |
c92ff1bd | 427 | |
c1a2f7f0 RR |
428 | /* |
429 | * Special counters, in some configurations protected by the | |
430 | * page_table_lock, in other configurations by being atomic. | |
431 | */ | |
432 | struct mm_rss_stat rss_stat; | |
801460d0 | 433 | |
c1a2f7f0 | 434 | struct linux_binfmt *binfmt; |
6345d24d | 435 | |
c1a2f7f0 RR |
436 | /* Architecture-specific MM context */ |
437 | mm_context_t context; | |
c92ff1bd | 438 | |
c1a2f7f0 | 439 | unsigned long flags; /* Must use atomic bitops to access */ |
c92ff1bd | 440 | |
c1a2f7f0 | 441 | struct core_state *core_state; /* coredumping support */ |
a961e409 | 442 | #ifdef CONFIG_MEMBARRIER |
c1a2f7f0 | 443 | atomic_t membarrier_state; |
a961e409 | 444 | #endif |
858f0993 | 445 | #ifdef CONFIG_AIO |
c1a2f7f0 RR |
446 | spinlock_t ioctx_lock; |
447 | struct kioctx_table __rcu *ioctx_table; | |
858f0993 | 448 | #endif |
f98bafa0 | 449 | #ifdef CONFIG_MEMCG |
c1a2f7f0 RR |
450 | /* |
451 | * "owner" points to a task that is regarded as the canonical | |
452 | * user/owner of this mm. All of the following must be true in | |
453 | * order for it to be changed: | |
454 | * | |
455 | * current == mm->owner | |
456 | * current->mm != mm | |
457 | * new_owner->mm == mm | |
458 | * new_owner->alloc_lock is held | |
459 | */ | |
460 | struct task_struct __rcu *owner; | |
78fb7466 | 461 | #endif |
c1a2f7f0 | 462 | struct user_namespace *user_ns; |
925d1c40 | 463 | |
c1a2f7f0 RR |
464 | /* store ref to file /proc/<pid>/exe symlink points to */ |
465 | struct file __rcu *exe_file; | |
cddb8a5c | 466 | #ifdef CONFIG_MMU_NOTIFIER |
c1a2f7f0 | 467 | struct mmu_notifier_mm *mmu_notifier_mm; |
e7a00c45 | 468 | #endif |
e009bb30 | 469 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS |
c1a2f7f0 | 470 | pgtable_t pmd_huge_pte; /* protected by page_table_lock */ |
cbee9f88 PZ |
471 | #endif |
472 | #ifdef CONFIG_NUMA_BALANCING | |
c1a2f7f0 RR |
473 | /* |
474 | * numa_next_scan is the next time that the PTEs will be marked | |
475 | * pte_numa. NUMA hinting faults will gather statistics and | |
476 | * migrate pages to new nodes if necessary. | |
477 | */ | |
478 | unsigned long numa_next_scan; | |
cbee9f88 | 479 | |
c1a2f7f0 RR |
480 | /* Restart point for scanning and setting pte_numa */ |
481 | unsigned long numa_scan_offset; | |
6e5fb223 | 482 | |
c1a2f7f0 RR |
483 | /* numa_scan_seq prevents two threads setting pte_numa */ |
484 | int numa_scan_seq; | |
20841405 | 485 | #endif |
c1a2f7f0 RR |
486 | /* |
487 | * An operation with batched TLB flushing is going on. Anything | |
488 | * that can move process memory needs to flush the TLB when | |
489 | * moving a PROT_NONE or PROT_NUMA mapped page. | |
490 | */ | |
491 | atomic_t tlb_flush_pending; | |
3ea27719 | 492 | #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH |
c1a2f7f0 RR |
493 | /* See flush_tlb_batched_pending() */ |
494 | bool tlb_flush_batched; | |
6345d24d | 495 | #endif |
c1a2f7f0 | 496 | struct uprobes_state uprobes_state; |
5d317b2b | 497 | #ifdef CONFIG_HUGETLB_PAGE |
c1a2f7f0 | 498 | atomic_long_t hugetlb_usage; |
5d317b2b | 499 | #endif |
c1a2f7f0 | 500 | struct work_struct async_put_work; |
133ff0ea JG |
501 | |
502 | #if IS_ENABLED(CONFIG_HMM) | |
c1a2f7f0 RR |
503 | /* HMM needs to track a few things per mm */ |
504 | struct hmm *hmm; | |
133ff0ea | 505 | #endif |
c1a2f7f0 RR |
506 | } __randomize_layout; |
507 | ||
508 | /* | |
509 | * The mm_cpumask needs to be at the end of mm_struct, because it | |
510 | * is dynamically sized based on nr_cpu_ids. | |
511 | */ | |
512 | unsigned long cpu_bitmap[]; | |
513 | }; | |
c92ff1bd | 514 | |
abe722a1 IM |
515 | extern struct mm_struct init_mm; |
516 | ||
c1a2f7f0 | 517 | /* Pointer magic because the dynamic array size confuses some compilers. */ |
6345d24d LT |
518 | static inline void mm_init_cpumask(struct mm_struct *mm) |
519 | { | |
c1a2f7f0 RR |
520 | unsigned long cpu_bitmap = (unsigned long)mm; |
521 | ||
522 | cpu_bitmap += offsetof(struct mm_struct, cpu_bitmap); | |
523 | cpumask_clear((struct cpumask *)cpu_bitmap); | |
6345d24d LT |
524 | } |
525 | ||
45e575ab | 526 | /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */ |
de03c72c KM |
527 | static inline cpumask_t *mm_cpumask(struct mm_struct *mm) |
528 | { | |
c1a2f7f0 | 529 | return (struct cpumask *)&mm->cpu_bitmap; |
de03c72c | 530 | } |
45e575ab | 531 | |
56236a59 MK |
532 | struct mmu_gather; |
533 | extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, | |
534 | unsigned long start, unsigned long end); | |
535 | extern void tlb_finish_mmu(struct mmu_gather *tlb, | |
536 | unsigned long start, unsigned long end); | |
537 | ||
16af97dc | 538 | static inline void init_tlb_flush_pending(struct mm_struct *mm) |
20841405 | 539 | { |
16af97dc | 540 | atomic_set(&mm->tlb_flush_pending, 0); |
20841405 | 541 | } |
16af97dc NA |
542 | |
543 | static inline void inc_tlb_flush_pending(struct mm_struct *mm) | |
20841405 | 544 | { |
16af97dc | 545 | atomic_inc(&mm->tlb_flush_pending); |
af2c1401 | 546 | /* |
8b1b436d PZ |
547 | * The only time this value is relevant is when there are indeed pages |
548 | * to flush. And we'll only flush pages after changing them, which | |
549 | * requires the PTL. | |
550 | * | |
551 | * So the ordering here is: | |
552 | * | |
040cca3a | 553 | * atomic_inc(&mm->tlb_flush_pending); |
8b1b436d PZ |
554 | * spin_lock(&ptl); |
555 | * ... | |
556 | * set_pte_at(); | |
557 | * spin_unlock(&ptl); | |
558 | * | |
559 | * spin_lock(&ptl) | |
560 | * mm_tlb_flush_pending(); | |
561 | * .... | |
562 | * spin_unlock(&ptl); | |
563 | * | |
564 | * flush_tlb_range(); | |
040cca3a | 565 | * atomic_dec(&mm->tlb_flush_pending); |
8b1b436d | 566 | * |
0e709703 PZ |
567 | * Where the increment if constrained by the PTL unlock, it thus |
568 | * ensures that the increment is visible if the PTE modification is | |
569 | * visible. After all, if there is no PTE modification, nobody cares | |
570 | * about TLB flushes either. | |
571 | * | |
572 | * This very much relies on users (mm_tlb_flush_pending() and | |
573 | * mm_tlb_flush_nested()) only caring about _specific_ PTEs (and | |
574 | * therefore specific PTLs), because with SPLIT_PTE_PTLOCKS and RCpc | |
575 | * locks (PPC) the unlock of one doesn't order against the lock of | |
576 | * another PTL. | |
577 | * | |
578 | * The decrement is ordered by the flush_tlb_range(), such that | |
579 | * mm_tlb_flush_pending() will not return false unless all flushes have | |
580 | * completed. | |
af2c1401 | 581 | */ |
20841405 | 582 | } |
16af97dc | 583 | |
16af97dc | 584 | static inline void dec_tlb_flush_pending(struct mm_struct *mm) |
20841405 | 585 | { |
0a2c4048 | 586 | /* |
0e709703 PZ |
587 | * See inc_tlb_flush_pending(). |
588 | * | |
589 | * This cannot be smp_mb__before_atomic() because smp_mb() simply does | |
590 | * not order against TLB invalidate completion, which is what we need. | |
591 | * | |
592 | * Therefore we must rely on tlb_flush_*() to guarantee order. | |
0a2c4048 | 593 | */ |
16af97dc | 594 | atomic_dec(&mm->tlb_flush_pending); |
20841405 | 595 | } |
20841405 | 596 | |
0e709703 PZ |
597 | static inline bool mm_tlb_flush_pending(struct mm_struct *mm) |
598 | { | |
599 | /* | |
600 | * Must be called after having acquired the PTL; orders against that | |
601 | * PTLs release and therefore ensures that if we observe the modified | |
602 | * PTE we must also observe the increment from inc_tlb_flush_pending(). | |
603 | * | |
604 | * That is, it only guarantees to return true if there is a flush | |
605 | * pending for _this_ PTL. | |
606 | */ | |
607 | return atomic_read(&mm->tlb_flush_pending); | |
608 | } | |
609 | ||
610 | static inline bool mm_tlb_flush_nested(struct mm_struct *mm) | |
611 | { | |
612 | /* | |
613 | * Similar to mm_tlb_flush_pending(), we must have acquired the PTL | |
614 | * for which there is a TLB flush pending in order to guarantee | |
615 | * we've seen both that PTE modification and the increment. | |
616 | * | |
617 | * (no requirement on actually still holding the PTL, that is irrelevant) | |
618 | */ | |
619 | return atomic_read(&mm->tlb_flush_pending) > 1; | |
620 | } | |
621 | ||
f872f540 AL |
622 | struct vm_fault; |
623 | ||
624 | struct vm_special_mapping { | |
625 | const char *name; /* The name, e.g. "[vdso]". */ | |
626 | ||
627 | /* | |
628 | * If .fault is not provided, this points to a | |
629 | * NULL-terminated array of pages that back the special mapping. | |
630 | * | |
631 | * This must not be NULL unless .fault is provided. | |
632 | */ | |
a62c34bd | 633 | struct page **pages; |
f872f540 AL |
634 | |
635 | /* | |
636 | * If non-NULL, then this is called to resolve page faults | |
637 | * on the special mapping. If used, .pages is not checked. | |
638 | */ | |
b3ec9f33 SJ |
639 | vm_fault_t (*fault)(const struct vm_special_mapping *sm, |
640 | struct vm_area_struct *vma, | |
641 | struct vm_fault *vmf); | |
b059a453 DS |
642 | |
643 | int (*mremap)(const struct vm_special_mapping *sm, | |
644 | struct vm_area_struct *new_vma); | |
a62c34bd AL |
645 | }; |
646 | ||
d17d8f9d DH |
647 | enum tlb_flush_reason { |
648 | TLB_FLUSH_ON_TASK_SWITCH, | |
649 | TLB_REMOTE_SHOOTDOWN, | |
650 | TLB_LOCAL_SHOOTDOWN, | |
651 | TLB_LOCAL_MM_SHOOTDOWN, | |
5b74283a | 652 | TLB_REMOTE_SEND_IPI, |
d17d8f9d DH |
653 | NR_TLB_FLUSH_REASONS, |
654 | }; | |
655 | ||
bd6dace7 TH |
656 | /* |
657 | * A swap entry has to fit into a "unsigned long", as the entry is hidden | |
658 | * in the "index" field of the swapper address space. | |
659 | */ | |
660 | typedef struct { | |
661 | unsigned long val; | |
662 | } swp_entry_t; | |
663 | ||
5b99cd0e | 664 | #endif /* _LINUX_MM_TYPES_H */ |